US AI Models Outpace Chinese Counterparts by an Average of 7 Months Since 2023

Since 2023, the United States has consistently developed models at the forefront of AI capabilities, as measured by the Epoch Capabilities Index (ECI). Over this period, Chinese models have lagged behind their US counterparts by an average of seven months, with a minimum gap of four months and a maximum of 14 months. This trend closely mirrors the divide between proprietary and open-weight models, given that most leading Chinese models are open-source, while top US models remain closed.

Key Findings

• US Leadership: Every model at the cutting edge of AI capabilities has been developed in the United States since 2023.
• Chinese Lag: Chinese models have trailed by an average of seven months, with a range from four to 14 months.
• Open vs. Proprietary: The gap aligns with the broader disparity between open-source and proprietary models.

Analysis

To understand the capabilities gap, we analyzed the ECI scores of leading US and Chinese models over time. Here’s how it was calculated:

• Model Selection: We identified the set of models that had the highest ECI upon release for their respective countries.
• Initial Models Excluded: The first models from each country (LLaMA-65B for the US and Baichuan1-7B for China) were excluded to ensure a fair comparison, as they might not have been at the true frontier of capabilities.
• Daily Gap Calculation: For each day, we determined the ECI of the best Chinese model and calculated how long it had been since the leading US model was at or below that score. Models are considered equivalent if their scores differ by less than one ECI point.

Notable Data Points

• May 2024 Milestone: The first instance where a Chinese model (Qwen3 Max) surpassed GPT-4's ECI, marking a 14-month gap.
• April 2025 Benchmark: As of April 2025, no Chinese model has yet matched the ECI of OpenAI’s o3 model.

Visualizing the Gap

The graph below illustrates the capabilities gap between US and Chinese models over time. Each point represents the number of months by which Chinese models lag behind their US counterparts.

|CC-BY epoch.aiDownload graph

Related Insights

• US vs. Non-US Performance: Updated Feb. 7, 2025
  • US models currently outperform non-US models across various benchmarks.
• Chinese Compute Trends: Jan. 22, 2025
  • Chinese language models have scaled up more slowly compared to their global counterparts.
• AI Supercomputers Performance: Jun. 5, 2025
  • The US hosts the majority of GPU cluster performance, followed by China.

Implications for Practitioners

The persistent gap in AI capabilities between the US and China has several implications:

• Research and Development: US researchers have access to more advanced models, potentially accelerating innovation.
• Open Source Ecosystem: Chinese developers rely heavily on open-source models, which may limit their ability to push the boundaries of AI.
• Geopolitical Dynamics: The gap could influence global tech policies and collaborations.

Conclusion

The data from Epoch AI highlights a significant and consistent lead for US-developed AI models since 2023. This trend underscores the importance of proprietary development in maintaining a competitive edge in AI capabilities. As the field continues to evolve, it will be crucial to monitor how these dynamics shift and what new strategies emerge.